There was a time when the economics of a collection agency were pretty simple. Measure how many accounts one collector could effectively handle: Accounts per Collector, or ACR. The higher the number, the lower the cost to collect.

Collections was primarily a telephone business. Agents dialed phones and talked. Wrap-up time—writing notes, setting arrangements—was time away from the phone and therefore a target for productivity improvement. We measured the KPIs that mattered: average call time, wrap time, contacts, promises, and, of course, dollars collected. Perhaps the most concise measurement was revenue per seat.

Collectors were compensated a lot like salespeople. They were required to hit minimum collection quotas and earned progressively higher bonuses above them. We celebrated milestones. Collect a million dollars? There might be featured recognition, a jacket, and attendance at the annual achievers’ club dinner.

Collections was an agent-productivity game. Then came automation. Slowly at first. Automated dialers replaced manual dialing, although for a time regulations still required human “clickers” to initiate calls. Agents spent less time making cold calls and more time taking productive ones. Letters became more targeted, moving beyond required initial notices. Then came email. Then texting. A text message costs a fraction of a mailed letter. Outreach that once consumed postage, paper, printing, and labor suddenly cost almost nothing.

Regulation increasingly became the constraint. Controls were required to enforce calling-frequency and consent rules.

But economically, something important was happening. Technology was still being evaluated primarily by one standard: Did it make the collector more productive?

And it did. Spectacularly.

We became so good at generating contacts that eventually we created another problem. We could drive customers to call us faster than our agents could answer the phones.

The bottleneck moved.

The Contact Center Leads the Way

Collections wasn’t alone in confronting that problem. Much of the innovation that followed came from the much larger contact-center industry. Banks, airlines, retailers, telecommunications companies, and insurers all faced versions of the same economic problem: employ armies of people to answer customer calls and labor quickly becomes the dominant cost.

The solutions evolved. Calls were routed more efficiently. Labor moved offshore and nearshore. Voice automation managed queues and increasingly handled simple transactions itself. But each improvement exposed another constraint. Generate a customer response and then lose that customer in a queue, and the money spent creating the contact has been wasted.

So the race accelerated. Voice recognition improved. Natural-language processing improved. Conversational AI began replacing the rigid instruction trees customers hated.

And the KPIs evolved alongside the technology:

• Average handle time

• Dropped calls

• First-call resolution

• Containment

The question was no longer simply how efficiently an agent could handle an interaction. It became: How much of the interaction could be completed without reaching a human being at all?

Every improvement attacked the largest cost in the contact center: labor.

Collections followed Contact Centers. Why wouldn’t we? We had agents answering phones too, and the technology worked.

For the first time, we could calculate something resembling ACR for the bot. The number was staggering. A machine could “handle” a volume of accounts that made traditional accounts-per-collector measurements almost meaningless.

So the KPIs moved with the bottleneck. We moved from accounts per collector toward contact rates, dropped calls, containment, digital conversion, portal utilization, and ultimately payment completion.

Every successfully contained call wasn’t merely a more productive agent interaction. There was no agent interaction.The labor cost disappeared. That distinction matters. Technology was no longer simply making the unit of production more productive.

Technology was becoming the unit of production.

The remaining agents didn’t disappear. They became different. They increasingly handled complicated calls: higher balances, unusual circumstances, disputes, emotionally delicate conversations, and, yes, customers who simply wanted another human being.

Measuring those agents against the old ACR or dollars-per-seat standards misses the point. We intentionally routed the easiest transactions away from them.

Meanwhile, customer-facing technology kept improving. Texts substituted for letters. Email performance could be measured by delivery and response. Dialing strategies could be refined using customer and debt-class characteristics. Right-party contacts, digital responses, portal conversions, and payment completion became increasingly important indicators.

The customer transaction itself became more frictionless. Checks gave way to debit cards, credit cards, ACH, wires, Apple Pay, Google Pay, Venmo, and other options.

Make it easy for the customer.

The results were real. Margins improved. Agencies without the capital to invest in increasingly efficient outreach technology fell behind. The changes encouraged consolidation. Higher margins attracted competitors, while clients increasingly expected a share of the efficiencies through pricing.

But the success of customer-facing technology in collections comes with a twist. A contact center and a collection agency can look remarkably similar from the front: rows of agents with headsets, taking and making calls, having customer conversations while supervisors watch dashboards. Walk through the back door, though, and they are very different businesses.

A traditional contact center primarily manages interactions. A collection agency also manages money and balances. That’s not a small distinction.

A customer calling an airline wants to change a flight. A customer calling a retailer may want to return a product. A customer calling a collection agency may not even know the name of the company answering the phone.

We first have to determine who the customer is. Then determine which obligation we’re discussing. Then make the required disclosures, establish the current balance, perhaps negotiate a settlement, create a payment plan, accept a payment, post it against the correct obligation, calculate the agency’s fee, place the creditor’s money into the appropriate trust account, remit it to the client, and report it.

Then, all too frequently, we reverse some or all of that when a payment comes back NSF. Throughout the process, the creditor maintains its own records of the same debt, mirroring what the agency has.

Collections isn’t merely a contact-center business. It is a financial transaction-processing business wearing a contact-center headset.

I’m not sure we fully appreciated that distinction as the technology race accelerated. Why would we? The results were spectacular. Automated outreach increased contacts. Conversational AI increased containment. Portals increased self-service. Digital payment options increased conversion.

Every dashboard told us the same thing: Faster. Cheaper. Better.Investment naturally chased the visible constraint: the customer.And perhaps that created some myopia.

The contact-center technology pioneers understandably attacked the problems they knew best. Collection agencies understandably wanted those tools as quickly as they could get them.

Meanwhile, sitting underneath all of it was the old collection core. The foundation.

It isn’t exciting. It isn’t conversational. Nobody is demonstrating trust-account reconciliation on a splashy conference stage. It just keeps track of the money.

So we kept building around it. Another interface. Another client customization. Another payment type. Another API. Another patch. Another workaround. The irony is that every successful customer-facing innovation increased the burden placed upon the core. More contacts created more arrangements. More arrangements created more payments. More payment methods created more transaction types. More transactions created more exceptions. More automation demanded faster and more reliable synchronization.

We were accelerating the front of the factory while leaving the machinery at the back largely untouched. Eventually, the bottleneck had to move there.

And replacing that infrastructure is much harder than adding another customer-facing tool. Legacy collection systems carry history: client-specific customizations, debt-specific workflows, payment rules, regulatory requirements, interfaces, APIs, reporting conventions, and years of patches layered over patches. Replacing the core isn’t installing another bot. It requires understanding the collection process end to end. And it is expensive.

That creates an uncomfortable economic squeeze. At precisely the moment clients are demanding their share of productivity gains through lower contingency rates, agencies need significant capital investment to replace the infrastructure necessary to deliver those gains reliably. Prices down. Investment up.

There is another economic change underneath that squeeze. The industry’s cost structure is shifting. The traditional collection agency scaled largely by adding people. Labor was a major variable cost, and productivity meant getting more output from each collector.

Technology changes that equation. Modern platforms, AI, cybersecurity, compliance systems, data infrastructure, integrations, and transaction-processing systems require substantial investment, but once built, the marginal cost of processing another automated interaction can be very small. The economics therefore increasingly reward scale.

That almost certainly accelerates consolidation. The mom-and-pop collection agency increasingly resembles the mom-and-pop coffee shop competing against Dunkin’. Can it compete? Certainly. But without a defensible niche, competing with the technology investment, infrastructure, and economies of scale of larger operators becomes increasingly difficult.

Perhaps the future small agency looks less like an independent operator and more like a specialist plugged into somebody else’s technology platform.

And that leads to a more fundamental question. Why did creditors outsource collections in the first place?

The answer was economically straightforward. A creditor had already written off the debt. Building and maintaining the people, facilities, dialing infrastructure, compliance organization, payment systems, and management required to pursue those accounts internally often made little sense. Give the accounts to a specialist. Pay a contingency fee. Buy rather than build was an easy decision. Except that equation is changing.

Put automated outreach, conversational AI, digital portals, automated payment processing, and intelligent account segmentation directly in the hands of the creditor and collecting internally no longer necessarily requires building a giant collection floor filled with agents.

There is another advantage rarely discussed outside the back offices of creditors and their agencies. The interface disappears.Anyone who has worked there understands the problem. Creditor and agency maintain two systems describing the same debt. Balances move back and forth. Payments arrive at either party. Payments reverse. Customers negotiate. Accounts are recalled. Files are transmitted. Transactions have to remain synchronized.

Two independent databases spend enormous effort agreeing about one customer’s obligation.

Remove the intermediary and much of that complexity disappears with it.

So perhaps the ultimate question isn’t how much more productive AI can make a collection agency. It is whether the same technology eventually changes the economic reason for having the intermediary at all.

I don’t think collection agencies disappear. Creditors aren’t outsourcing only collector labor. They are also buying specialization, licensing infrastructure, regulatory expertise, compliance management, variable capacity, technology, benchmarking, litigation-risk management, and management focus. For many creditors, building all of that internally will continue to make little economic sense.

But that leads to a different question:

What comparative advantage remains with the agency when labor intensity is no longer its dominant advantage?

The answer may reshape the industry. Large creditors may bring more collections inside. Large agencies may increasingly become technology platforms with collection expertise attached. Smaller creditors may outsource precisely because they cannot afford the infrastructure themselves. Specialized agencies may survive because expertise in particular asset classes, customers, or regulatory environments remains difficult to replicate. The traditional labor-intensive agency in the middle gets squeezed.

Could we have anticipated some of this? Probably. And this may be where the story becomes about more than collections.

Technology development celebrates the new for good reason. Legacy systems are frustrating. They’re slow. They’re patched together. Their architecture can look incomprehensible to someone designing with modern tools. The temptation is to look at that complexity and conclude that previous generations simply designed badly. Sometimes they did. But sometimes complexity is accumulated knowledge. That strange process exists because something once broke. That ugly exception handles the transaction nobody anticipated. That redundant reconciliation exists because two systems once disagreed about who owned the money.

The people operating those systems often understand things that aren’t obvious from the architecture diagram. Perhaps we should listen to them more carefully. Not because we should slow innovation. And certainly not because legacy infrastructure should be preserved forever.

Quite the opposite.

Had we understood earlier where the dependencies really lived, perhaps some of the enormous investment flowing into the customer-facing revolution could have been balanced with investment in the foundation underneath it. We might have moved slightly slower at the front.

And arrived faster at the end.

That seems worth remembering now because AI is accelerating technology development far beyond collections. The pressure everywhere is to move faster. Replace. Automate. Eliminate friction. Tear out the legacy. Much of that will be exactly right.But before dismissing the old operator defending an ugly process, perhaps ask one more question: Why is it there?

There may be thirty years of accumulated technical debt. There may also be thirty years of accumulated knowledge. Knowing the difference may become considerably more important as the speed of change accelerates.

For decades we measured the productivity of the collector because the collector was the scarce resource. Technology changed that. Then customer contact became the constraint.Technology changed that. Now the constraint is migrating into the infrastructure underneath the transaction. Technology will change that too. And eventually technology may alter the economics of the intermediary itself.

Every innovation solves yesterday’s bottleneck. Then exposes the next one.

The bottleneck moved. The economics followed it. And they’re still moving.